[PATCH] D111533: [SCEV] Invalidate user SCEVs along with operand SCEVs to avoid cache corruption

[Philip Reames via Phabricator via llvm-commits]

reames added a comment.

In D111533#3078423 <https://reviews.llvm.org/D111533#3078423>, @nikic wrote:

> (snip)

I generally agree that once we have users, we should revisit a bunch of how we do invalidation.  I have no problem with this being used in one place first, but I agree we can and should push this further.

> I would prefer it if we initially only landed the infrastructure to track users (without touching invalidation) and then migrate over individual invalidation users.

I have no problem with this approach.  I don't advocate it, but I'd be completely fine with the SCEV user infrastructure landing on it's own.

(Oh, one thing that occurred to me.  We should probably add verification/asserts of the use links themselves at some point.  I would be fine with that being a separate patch.)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D111533/new/

https://reviews.llvm.org/D111533